update OCR

src/inference.py

@@ -185,35 +185,50 @@ def ocr_note(img_path, backend="paddle"):
     if img is None:
         return ""
     
-    # Tiền xử lý mạnh: tăng độ phân giải, lọc nhiễu, làm sắc nét
+    # Tiền xử lý mạnh
     h, w = img.shape[:2]
     if w < 1000:
         scale = 1200 / w
         img = cv2.resize(img, (int(w*scale), int(h*scale)), interpolation=cv2.INTER_CUBIC)
     
     gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-    # Sử dụng CLAHE mạnh hơn
     clahe = cv2.createCLAHE(clipLimit=3.5, tileGridSize=(4,4))
     gray = clahe.apply(gray)
-    # Khử nhiễu nhưng giữ cạnh
     gray = cv2.bilateralFilter(gray, 9, 75, 75)
-    # Làm sắc nét
     kernel_sharpen = np.array([[-1,-1,-1],[-1,9,-1],[-1,-1,-1]])
     gray = cv2.filter2D(gray, -1, kernel_sharpen)
-    # Nhị phân hóa thích ứng
     gray = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
     img_proc = cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)
     
     reader = get_ocr_reader(backend)
-    if hasattr(reader, 'ocr'):
+    if hasattr(reader, 'ocr'):   # PaddleOCR
         result = reader.ocr(img_proc, cls=True)
         if result and result[0]:
             texts = [line[1][0].strip() for line in result[0] if line[1][1] > 0.3]
             return "\n".join(texts)
-    else:
-        results = reader.readtext(img_proc, detail=1, paragraph=True)
-        texts = [text for (_, text, conf) in results if conf > 0.3]
-        return "\n".join(texts)
+    else:                         # EasyOCR
+        results = reader.readtext(img_proc, detail=1, paragraph=False)
+        if not results:
+            return ""
+        # Sắp xếp theo tọa độ y (dòng) và x (cột)
+        results_sorted = sorted(results, key=lambda x: (x[0][0][1], x[0][0][0]))
+        lines = []
+        current_line = []
+        prev_y = None
+        for (pts, text, conf) in results_sorted:
+            if conf < 0.3:
+                continue
+            y_center = sum(p[1] for p in pts) / 4
+            if prev_y is None or abs(y_center - prev_y) < 15:
+                current_line.append(text.strip())
+            else:
+                if current_line:
+                    lines.append(" ".join(current_line))
+                current_line = [text.strip()]
+            prev_y = y_center
+        if current_line:
+            lines.append(" ".join(current_line))
+        return "\n".join(lines)
     return ""
 
 import cv2